Plasma membrane processes are differentially regulated by type I phosphatidylinositol phosphate 5-kinases and RASSF4

This paper explores regulation of membrane processes by phosphoinositide kinases. Phosphatidylinositol 4,5-bisphosphate [PtdIns(4,5)P₂], which constitutes <1% of total membrane phospholipids, is an essential cofactor for many plasma membrane proteins and a regulator of many membrane-associated processes (Balla, 2013; Dickson and Hille, 2019; Hille et al., 2015; Tan et al., 2015). It is produced primarily by type I phosphatidylinositol phosphate 5-kinases (PIP5KIs) (Choi et al., 2015; Loijens et al., 1996; Padrón et al., 2003; Wang et al., 2004), which phosphorylate the myo-inositol 5-hydroxyl of phosphatidylinositol 4-phosphate [PtdIns(4)P]. Three distinct genes for PIP5KIs have been described: PIP5KIA, PIP5KIB and PIP5KIC, which encode isoforms PIP5KIα, PIP5KIβ and PIP5KIγ, respectively (Tuosto et al., 2015). Northern blot analysis has shown that PIP5KIα predominates in skeletal muscle, PIP5KIβ in heart and PIP5KIγ in brain (Ishihara et al., 1998; Loijens and Anderson, 1996). Here, we investigate the functional specificity of the β and γ isoforms. They can form homo- and heterodimers that enlarge the bilayer interaction interface and enhance their activity (Lacalle et al., 2015). PIP5KIβ and PIP5KIγ localize primarily to the plasma membrane, often with accessory proteins (Lacalle et al., 2015; Padrón et al., 2003; Wang et al., 2004). The literature also gives each kinase isoform additional minor and distinct subcellular distributions; for example, PIP5KIβ is reported in perinuclear vesicles and PIP5KIγ in focal adhesions and endosomes (Choi et al., 2015; Tuosto et al., 2015). Small differences in distribution might translate into distinct functions. Thus, siRNA experiments in HeLa cells have identified the short splice variant of PIP5KIγ as the major contributor to the PtdIns(4,5)P₂ pool that supports generation of inositol trisphosphate (InsP₃) mediated by G protein-coupled receptors (Wang et al., 2004). In megakaryocytes, PIP5KIγ^−/− cells show decreased association between the plasma membrane and the cytoskeleton, a phenotype that could be rescued by overexpression of PIP5KIγ but not PIP5KIβ (Wang et al., 2008). Furthermore, in T cells, PIP5KIβ is specifically recruited to the immune synapse, producing PtdIns(4,5)P₂ pools necessary for T-cell activation (Kallikourdis et al., 2016).

Here, we compare the efficacy of PIP5KI enzymes on two PtdIns(4,5)P₂-sensitive processes: activation of KCNQ2/3 (K_V7.2/7.3) channels and the steps leading to store-operated calcium entry (SOCE). KCNQ2/3 channels conduct an outwardly rectified potassium current that regulates the subthreshold excitability of sympathetic and central neurons (Brown and Adams, 1980; Wang et al., 1998). Regulation of KCNQ2/3 channels by PtdIns(4,5)P₂ has been widely studied (Falkenburger et al., 2010; Li et al., 2005; Suh and Hille, 2002; Zhang et al., 2003). PtdIns(4,5)P₂ increases the open probability of channels made from KCNQ subunits (Li et al., 2005), and strong PtdIns(4,5)P₂ depletion reduces KCNQ2/3 channel current reversibly and almost completely (Suh and Hille, 2002; Zhang et al., 2003). Rapid lipid depletion can be accomplished via activation of G_q-coupled membrane receptors (including M₁ muscarinic receptors) or artificially via chemical recruitment of a lipid phosphatase or activation of a voltage-sensitive lipid phosphatase (VSP) (Falkenburger et al., 2010; Murata et al., 2005). KCNQ channels are good on-line indicators of dynamic changes in PtdIns(4,5)P₂ pools at the plasma membrane.

SOCE is the main mechanism for refilling the endoplasmic reticulum (ER) with calcium in non-excitable cells. SOCE begins as the ER membrane-resident protein stromal interaction molecule 1 (STIM1) senses a reduction in ER luminal calcium, oligomerizes, translocates to ER–plasma membrane junctions forming puncta and transactivates plasma membrane Orai calcium channels (Liou et al., 2007; Luik et al., 2006; Wu et al., 2006). It is widely agreed that phosphoinositides enhance translocation of STIM1 and development of SOCE, although they might not be absolutely essential. The polybasic, lysine-rich tail of STIM1 associates with negatively charged phosphoinositides in the plasma membrane, with preference for PtdIns(4,5)P₂ (Ercan et al., 2009). STIM1 without its lysine-rich tail (STIM1-ΔK) failed to form puncta after store depletion when overexpressed alone, but coexpression of excess calcium release-activated calcium channel protein 1 (Orai1) was sufficient to restore thapsigargin-stimulated puncta formation and SOCE (Park et al., 2009). Reductions in STIM1 translocation or of calcium entry through Orai1 channels have been reported in response to simultaneous pharmacological inhibition of lipid 3- and 4-kinases (Korzeniowski et al., 2009; Walsh et al., 2010), knockout of the lipid 4-kinase PI4KIIIα (Nakatsu et al., 2012) or recruitment of a PtdIns(4,5)P₂ 5-phosphatase to the plasma membrane, especially when combined with inhibition of 3- and 4-kinases (Korzeniowski et al., 2009; Walsh et al., 2010). These manipulations usually reduce more than one species of phosphoinositide simultaneously. Taken together, such studies endorse PtdIns(4)P, PtdIns(4,5)P₂ and PtdIns(3,4,5)P₃ as positive cofactors for STIM1 translocation to plasma membrane junctions.

In addition to probing PtdIns(4,5)P₂ signaling specificity through two PIP5K isoforms, we investigated a protein that has been suggested to upregulate endogenous PIP5K enzyme activity. A human siRNA screen identified many proteins regulating SOCE, including a member of the Ras association domain family, RASSF4 (Chen et al., 2017). RASSF4 was found to promote STIM1 translocation to the plasma membrane, increase ER–plasma membrane junctions and upregulate total cellular PtdIns(4,5)P₂ in a manner dependent on 5-kinases. RASSF4 may stimulate phosphorylation of PtdIns(4)P indirectly by virtue of associating with ADP-ribosylation factor Arf6 (Chen et al., 2017), a positive regulator of PIP5KIβ (Honda et al., 1999) and PIP5KIγ (Krauss et al., 2003). RASSF4 is a tumor-suppressing protein of the Ras-effector family broadly expressed in human tissues (Eckfeld et al., 2004; Han et al., 2016; Zhang et al., 2017). It is thought to adjust the balance of Ras signaling away from cell proliferation and survival in favor of apoptosis (De Smedt et al., 2018).

In this study, we show divergent effects of two PIP5KI isoforms and RASSF4 on KCNQ2/3 currents and development of SOCE. Some specificity in PtdIns(4,5)P₂ signaling is encoded by the PIP5KI isozymes. We found that PIP5KIs do not confer specificity via production of distinct chemical species of PtdIns(4,5)P₂. Their divergent effects imply localization to distinct membranes or membrane domains and regulatory associations with different effectors.

We first studied localization of overexpressed PIP5KIβ, PIP5KIγ and RASSF4 by confocal microscopy and total internal reflection fluorescence (TIRF) microscopy. Plasmids were transiently transfected in tsA201 cells. PIP5KI enzymes catalyze the synthesis of PtdIns(4,5)P₂, the phosphoinositide that marks the plasma membrane. Consistent with that, confocal images showed that both PIP5KIβ and PIP5KIγ were localized mainly at the plasma membrane (Fig. 1A). TIRF microscopy corroborated the predominant expression of PIP5KIβ and PIP5KIγ in the plasma membrane, showing strong fluorescence in the TIRF footprint (Fig. 1B). For comparison, we also overexpressed RASSF4, a putative upstream activator of PIP5KI enzymes via Arf6 (Chen et al., 2017). RASSF4 showed diffuse intracellular expression in confocal images (Fig. 1A) and weak and out-of-focus fluorescence in TIRF (Fig. 1B). Z-stack analysis revealed no change in cell thickness (Fig. S1A), and electrical measurements of cell membrane capacitance revealed no statistically significant differences in cell membrane surface area (Fig. S1B). Nevertheless, cells overexpressing RASSF4 had an altered morphology with more extensive filopodia and lamellipodia (Fig. S1C-E).

We used two phosphatidylinositol bisphosphate (PtdInsP₂)-binding probes, PH-PLCδ1 and Tubby_C^R332H, to assess PtdInsP₂ distribution for each overexpression condition. PH-PLCδ1 can bind both PtdIns(4,5)P₂ (K_D ∼2 μM) and inositol (1,4,5)-trisphosphate (Ins(1,4,5)P₃; K_D ∼100 nM) (Wills et al., 2018) and translocates readily from the plasma membrane to the cytosol upon PtdIns(4,5)P₂ hydrolysis. Tubby_C^R332H has a much lower affinity for PtdIns(4,5)P₂ than does PH-PLCδ1 and can also bind PtdIns(3,4)P₂ and PtdIns(3,4,5)P₃ (Wills et al., 2018). These probes show the relative abundance of PtdIns(4,5)P₂ across the cell and are useful real-time indicators of cellular PtdIns(4,5)P₂ dynamics, but do not report absolute levels of PtdIns(4,5)P₂. In control cells, the PH-PLCδ1 probe with high affinity for PtdIns(4,5)P₂ distributed to the cell surface (Fig. 1C,D), whereas the Tubby_C^R332H probe with lower affinity for PtdIns(4,5)P₂ distributed significantly in the cytoplasm in addition to having clear localization at the cell surface (Fig. 1E,F). In cells overexpressing PIP5KIβ and PIP5KIγ, both probes were drawn strongly to the cell surface, suggesting a marked rise in PtdIns(4,5)P₂ at the plasma membrane. In contrast, in cells overexpressing RASSF4, both probes showed significantly less cell surface localization, even less than in the control (Fig. 1C-F). RASSF4 cells also showed an increased number of labeled intracellular bodies with Tubby_C^R332H (Fig. S1F-H). Confocal sections placed most of these bodies inside the nucleus (Fig. S1F). These results are consistent with RASSF4 shifting a fraction of cellular PtdInsP₂ to intracellular sites. To check for potentiation of PIP5KI enzymes by RASSF4, we coexpressed PIP5KIβ or PIP5KIγ with RASSF4 and assessed cellular PtdIns(4,5)P₂ distribution. Compared with cells containing each PIP5KI enzyme alone, coexpression with RASSF4 increased the plasma membrane component of Tubby_C^R332H in cells coexpressing PIP5KIγ, but made no difference in cells coexpressing PIP5KIβ (Fig. 1G,H). Interestingly, coexpression of PIP5KIβ and RASSF4 appeared to change nuclear morphology in the confocal sections, with 80% of cells containing two nuclei or a single nucleus with an unusual shape and/or invaginations as opposed to 30% of cells expressing RASSF4 alone or PIP5KIγ and RASSF4 (Fig. S1I).

The two PIP5KI enzymes, which had similar subcellular distributions at the resolution of conventional confocal and TIRF microscopy, also showed some functional similarities. We activated overexpressed M₁ muscarinic receptors by the extracellular agonist oxotremorine M (Oxo-M) and monitored the resulting depletion of plasma membrane PtdIns(4,5)P₂ from the time course of PH-PLCδ1 probe translocation to the cytosol. Translocation could be quantified from changes in the cytosolic intensity of PH-PLCδ1 (Fig. 1I). The translocation half-time was 17.7±0.9 s in control cells, 24.7±0.3 s in cells overexpressing PIP5KIβ and 21.3±0.7 s in cells overexpressing PIP5KIγ (Fig. 1K). The slower translocation of PH-PLCδ1 in PIP5KI-overexpressing cells is consistent with an expected enlarged basal plasma membrane PtdIns(4,5)P₂ pool that takes longer to deplete during receptor stimulation, combined with faster ongoing production of PtdIns(4,5)P₂ that could counteract breakdown by phospholipase C. The effect of PIP5KIβ was a little stronger than that of PIP5KIγ. In contrast, in cells overexpressing RASSF4, the translocation of PH-PLCδ1 to the cytosol was accelerated (half-time of 16.3±0.8 s in control cells and 12.5±1.0 s in cells overexpressing RASSF4; Fig. 1J,K) as if the basal plasma membrane pool of PtdIns(4,5)P₂ had been reduced by RASSF4 rather than increased. The half-time of PH-PLCδ1 return during PtdIns(4,5)P₂ recovery was not statistically different in cells overexpressing either of the 5-kinases compared with control cells, as if the 4-kinase activity (rather than the 5-kinases) is rate-limiting for recovery starting with phosphatidylinositol after long treatment with Oxo-M, in concordance with Falkenburger et al. (2010). The results so far are summarized in the first few lines of Table 1.

KCNQ2/3 ion channels require PtdIns(4,5)P₂ and are often used as rapid real-time indicators specific for plasma membrane PtdIns(4,5)P₂. We tested whether overexpressed PIP5KIβ, PIP5KIγ or RASSF4 regulated KCNQ2/3 channel function. Representative KCNQ2/3 current traces are shown in Fig. 2A for controls and cells overexpressing PIP5KIβ, PIP5KIγ or RASSF4 co-transfected with the Danio rerio voltage-sensitive lipid phosphatase (Dr-VSP). This enzyme rapidly removes the 5-phosphate from PtdIns(4,5)P₂ upon large depolarization of the membrane potential (Hossain et al., 2008; Murata et al., 2005). Partial dephosphorylation by Dr-VSP was evoked by a 1 s depolarization to +100 mV, reducing KCNQ current by 74±4% in control cells (Fig. 2A,B). Several observations suggested that overexpression of PIP5KIβ or PIP5KIγ but not RASSF4 enlarged the PtdIns(4,5)P₂ pools accessible to the channels and to Dr-VSP. The KCNQ2/3 current reduction by Dr-VSP was much reduced by overexpression of either 5-kinase (PIP5KIβ, 44±11%; PIP5KIγ, 41±12%; Fig. 2B) but not by overexpression of RASSF4 (71±5%; Fig. 2B). Similarly, recovery of currents representing refilling of the plasma membrane PtdIns(4,5)P₂ pool after brief activation of Dr-VSP was strongly accelerated by overexpression of either 5-kinase, especially PIP5KIγ (half-times: control, 8.4±0.6 s; PIP5KIβ, 3.8±1.3 s; PIP5KIγ, 1.4±0.4 s; Fig. 2C), but not by RASSF4 (9.7±0.5 s; Fig. 2C).

It has been reported that increasing basal PtdIns(4,5)P₂ in the plasma membrane accelerates voltage-dependent activation gating of KCNQ2/3 channels and slows deactivation gating (Dai et al., 2016). Hence, KCNQ2/3 current activation and deactivation kinetics reflect absolute levels of channel-available plasma membrane PtdIns(4,5)P₂. Accordingly, the rate of depolarization-induced activation gating of KCNQ2/3 currents was also accelerated by overexpression of PIP5KIβ or PIP5KIγ (exponential time constant, tau: control, 207±13 ms; PIP5KIβ, 150±12 ms: PIP5KIγ, 125±11 ms; Fig. 2D,E), whereas it was not statistically different for cells overexpressing RASSF4 compared with control cells (175±10 ms; Fig. 2D,E). Similarly, overexpression of PIP5KIβ and PIP5KIγ but not RASSF4 slowed deactivation of KCNQ2/3 currents (control, 68±4 ms; PIP5KIβ, 103±7 ms; PIP5KIγ, 110±8 ms; RASSF4, 52±5 ms; Fig. 2D,F). Again, overexpression of RASSF4 alone did not seem to augment the plasma membrane PtdIns(4,5)P₂ pool, whereas PIP5KIs did.

To check for potentiation of PIP5KI enzymes by RASSF4, we measured KCNQ2/3 currents in cells coexpressing 5-kinases and RASSF4. In comparison with cells overexpressing PIP5KIγ alone, coexpression of PIP5KIγ and RASSF4 reduced VSP inhibition of KCNQ2/3 current (control, 62±5%; PIP5KIγ, 18±6%; PIP5KIγ+RASSF4, 5±3%; Fig. 2G,I) and accelerated current recovery (Fig. 2G). Coexpression of PIP5KIγ and RASSF4 also accelerated KCNQ2/3 current activation (control, 196±13 ms; PIP5KIγ, 147±7 ms; PIP5KIγ+RASSF4, 112±8 ms; Fig. 2J,L) whereas deactivation was unchanged (Fig. 2J,M). Overexpression of RASSF4 together with PIP5KIβ did not change KCNQ2/3 current activation or deactivation compared with cells expressing PIP5KIβ alone (Fig. 2K-M). RASSF4 coexpression appears to potentiate production of plasma membrane PtdIns(4,5)P₂ by PIP5KIγ.

In short, both PIP5KIβ and PIP5KIγ regulate the kinetics of activation gating and deactivation gating of KCNQ2/3 channels and accelerate current recovery after Dr-VSP-mediated depletion of PtdIns(4,5)P₂. These results, summarized in Table 1, confirm that overexpression of PIP5KIβ or PIP5KIγ increases plasma membrane pools of PtdIns(4,5)P₂ that are accessible to plasma membrane KCNQ channels, PH-PLCδ1 probes, phospholipase C and Dr-VSP enzymes, and that they accelerate PtdIns(4,5)P₂ regeneration following Dr-VSP activity. These plasma membrane pools are not augmented by RASSF4 overexpression alone, but production of PtdIns(4,5)P₂ by PIP5KIγ is potentiated when RASSF4 and PIP5KIγ are coexpressed.

We completed the remaining experiments using RASSF4 without coexpression of PIP5KIs. Using only the endogenous kinases ensures unperturbed kinase expression and localization and no imposed bias for RASSF4 coupling to specific PIP5KI isoforms.

We turned next to a series of phosphoinositide-sensitive cellular events associated with SOCE. STIM1 proteins of the ER are mobilized to puncta at the plasma membrane where, in combination with Orai, they initiate channel opening and SOCE. PtdIns(4)P, PtdIns(4,5)P₂ and PtdIns(3,4,5)P₃ are positive regulators of this chain of events (Korzeniowski et al., 2009; Park et al., 2009; Saheki and De Camilli, 2017; Walsh et al., 2010). Unexpectedly, it was reported that, when overexpressed, the two kinases PIP5KIβ and PIP5KIγ regulate STIM1–Orai1 interactions oppositely: PIP5KIβ increases the Förster resonance energy transfer (FRET) between STIM1 and Orai1 as anticipated, whereas PIP5KIγ reduces the FRET (Calloway et al., 2011). Like PIP5KIβ, RASSF4 also stimulates formation of STIM1 puncta (Chen et al., 2017). We confirmed and extended these results by showing differential effects of PIP5KI isoforms at each step of SOCE. We considered successively the resting disposition of STIM1, its recruitment to near the plasma membrane by thapsigargin-induced emptying of calcium stores and the stimulated entry of calcium, SOCE. These steps involve the ER membrane as well as junctions with the plasma membrane. In TIRF microscopy, resting cells overexpressing RASSF4 had elevated basal STIM1 fluorescence in the cell footprint, whereas resting cells overexpressing PIP5KIγ had a low basal STIM1 fluorescence, like control cells (Fig. 3A,E). With PIP5KIβ, we observed some resting cells with only a few basal STIM1 puncta, like control cells, and some with many puncta (Fig. 3E). The cells with many STIM1 puncta were predominant, but thapsigargin failed to promote the formation of additional puncta in these cells, so they were excluded from the kinetic analysis described next.

The kinetic profiles of induction of STIM1 puncta formation by thapsigargin in TIRF microscopy were distinct for each PIP5KI or RASSF4 overexpression condition. PIP5KIβ overexpression accelerated STIM1 mobilization during thapsigargin compared with control cells, although the final net change was not larger (Fig. 3B-D). In contrast, PIP5KIγ accelerated STIM1 movement only a little but reduced the net change to just 35% of control. Puncta formation seemed supercharged by RASSF4 overexpression: more spontaneous puncta at rest and a boosted STIM1 response to thapsigargin. The percentage augmentation in STIM1 fluorescence by thapsigargin is given in Fig. 3B,C (control, 116±16%; PIP5KIβ, 119±22%; PIP5KIγ, 41±16%; RASSF4, 192±14%,) and the half-time of the increase in STIM1 fluorescence in Fig. 3B,D (control, 405±19 s; PIP5KIβ, 310±14 s; PIP5KIγ, 342±15 s; RASSF4, 204±28 s). We used MAPPER to label ER–plasma membrane junctions and found junctions more prominent in RASSF4-overexpressing cells compared with control cells and those overexpressing PIP5KIβ (Fig. 3F). Coexpression of STIM1 and MAPPER in RASSF4 cells elevated the basal fluorescence of both STIM1 and MAPPER even further (Fig. 3G). Differences in STIM1 distribution could also be observed in confocal microscopy. In cells overexpressing YFP-STIM1 alone, STIM1 fluorescence was predominant around the nucleus with some extensions toward the periphery of the cell as though it was diffusely present throughout the ER. In cells overexpressing PIP5KIβ, STIM1 fluorescence increased markedly in puncta near the plasma membrane in addition to a ring around the nucleus; with PIP5KIγ, the distribution was intermediate (Fig. 3H,I). In cells overexpressing RASSF4, YFP-STIM1 appeared to be excluded from the plasma membrane-proximal zone, with a strong ring around the nucleus in confocal images. Although the thin edges of the cell appeared to exclude STIM1 in confocal cross-sections, TIRF revealed STIM1 proximal to the plasma membrane in the cell's basal surface.

Differential regulation of STIM1 puncta formation by PIP5KIβ, PIP5KIγ and RASSF4 was only modestly paralleled by differential regulation of calcium entry through Orai1 channels. We measured intracellular calcium with the low-affinity calcium indicator Fura-4F. The cells were treated with 1 µM thapsigargin, first in 0 mM extracellular calcium to empty calcium stores followed by 0.5 mM calcium to allow SOCE (Fig. 4A). Cells overexpressing PIP5KIβ or RASSF4 had somewhat larger calcium increases than the control, whereas those with PIP5KIγ did not (Fig. 4B,D,E). The rate of calcium rise measured by the initial slope of the Fura-4F ratio was also increased in cells overexpressing PIP5KIβ or RASSF4, but not in cells overexpressing PIP5KIγ (Fig. 4B,C). To control for possible variability in the rate of perfusion, we also measured the rate of calcium increase after switching to 10 µM ionomycin in a 2 mM calcium solution; no differences among groups were detected. In addition to changes observed after thapsigargin, the application of 2 mM calcium before thapsigargin revealed a larger increase in intracellular calcium in cells overexpressing PIP5KIβ or RASSF4, consistent with our observation of increased formation of STIM1 puncta at rest. Although the kinetics and extent of SOCE varied somewhat with 5-kinase and RASSF4 overexpression, these changes were relatively modest compared with the large differences observed in STIM1 movement. SOCE appears to be a self-regulated process, despite changes in phosphoinositide levels and STIM1 puncta, consistent with the self-limiting calcium-dependent inhibition of Orai channels, which prevents overloading during calcium refilling (Zweifach and Lewis, 1995).

In summary, PIP5KIβ and RASSF4 bring STIM1 closer to the plasma membrane, accelerate STIM1 mobilization to the plasma membrane and lead to a faster calcium rise. PIP5KIγ does not bring STIM1 closer to the plasma membrane, reduces the amount of STIM1 that is recruited to the plasma membrane and does not change the rate of calcium rise. Although SOCE is influenced by phosphoinositides, the results we report do not parallel the observed changes in plasma membrane PtdIns(4,5)P₂ (see Table 1) and must be explained by additional assumptions invoking other membranes or other lipid and protein species.

We used mass spectrometry to resolve changes in total cellular phosphoinositide pools with overexpression of PIP5KIβ, PIP5KIγ and RASSF4. Our technique (Traynor-Kaplan et al., 2017) measures phosphoinositides in a population of cells collected from a culture dish. For each molecular weight species, the technique readily distinguishes total fatty acyl chain length (e.g. 34, 36, 38 or 40 carbons) and unsaturation (0-5 double bonds) as well as the number of phosphate groups on the inositol head group, but not their position on the ring [e.g. not PtdIns(3,5)P₂ versus PtdIns(4,5)P₂]. The measurements should be interpreted realizing that, with transiently transfected cells, only some cells in the population express the transfected plasmids at a significant level; thus, any mean changes observed are probably less than occurred in the subpopulation of well-transfected cells. We found, as might be anticipated, that total (unphosphorylated) phosphatidylinositol (PtdIns) content was unchanged in each of our kinase-upregulating conditions (Fig. 5A). PtdIns forms large mostly intracellular pools that are precursors of the phosphorylated forms. However, total phosphatidylinositol phosphate (PtdInsP, the substrate of PIP5Ks) was decreased by 15-25% (Fig. 5B), and total phosphatidylinositol bisphosphate (PtdInsP₂, the product of PIP5Ks) was increased by 10-40% (Fig. 5C). The change for PIP5KIβ-overexpressing cells was not statistically significant. Phosphatidylinositol trisphosphate (PtdInsP₃) increased by 30-220% (Fig. 5D). Together, these changes imply that PIP5KIβ, PIP5KIγ and RASSF4 each increase the net production of PtdInsP₂ and PtdInsP₃ from PtdInsP in the overall cellular phosphoinositide metabolism.

We also analyzed the fatty acid profile for each phosphoinositide type (Figs 5 and 6). Given the bulk changes in phosphoinositides by kinase overexpression or upregulation and the kinase-selective changes to plasma membrane PtdIns(4,5)P₂ and SOCE (Table 1), we tested whether the various kinases might confer specificity to PtdInsP₂-sensitive processes by differentially increasing particular chemical species of PtdInsP₂. In vitro, PIP5Ks can show some discrimination between PtdIns(4)P substrates based on their different fatty acyl chains (D'Souza and Epand, 2014; Muftuoglu et al., 2016; Shulga et al., 2012). However, we found that the acyl chain spectrum across species of PtdIns, PtdInsP, PtdInsP₂ and PtdInsP₃ was unchanged by overexpression of PIP5KIβ, PIP5KIγ or RASSF4 (expressed as nanograms per 10⁶ cells in Fig. 5 and as percentage of total PtdInsP₂ or PtdInsP₃ in Fig. 6A,B). Fig. 6C shows the percentage increment of specific PtdInsP₂ species compared with control. There were no statistically significant differences in the increments for different fatty acyl chains.

To further probe phosphoinositide involvement in the physiological activation of SOCE, we looked at stimulation of G protein-coupled receptors. Such receptors coupled to G_q activate phospholipase C to hydrolyze plasma membrane PtdIns(4,5)P₂, generating Ins(1,4,5)P₃, which triggers emptying of ER calcium stores and leads to SOCE. We have shown that PIP5Ks modulate SOCE, perhaps in part through changes in PtdIns(4)P and PtdIns(4,5)P₂ pools. So, we compared how weak versus strong depletion of PtdIns(4,5)P₂ via G_q-coupled receptors influences SOCE activation. We used Fura-4F as an indicator of intracellular calcium and FRET between PH-PLCδ1-CFP and PH-PLCδ1-YFP as an indicator of plasma membrane PtdIns(4,5)P₂. When M₁ receptors were overexpressed and extracellular calcium was absent, 1 µM Oxo-M treatment led to a modest intracellular calcium rise indicative of an emptying of intracellular stores. Upon addition of 2 mM extracellular calcium there was a second increase in calcium, indicative of SOCE, which was blocked by Orai channel pore blocker GSK-7975A (Fig. 7A).

It has been shown that whereas a high concentration (1 μM) of Oxo-M depletes PtdIns(4,5)P₂ in the plasma membrane, a low concentration (1 nM) still increases Ins(1,4,5)P₃ enough to activate sensitive Ins(1,4,5)P₃ receptors but without a detectable reduction in PtdIns(4,5)P₂ (Dickson et al., 2013). Because endogenous purinergic receptors are expressed at low density in these cells, the agonist UTP can also be used to generate Ins(1,4,5)P₃ and empty calcium stores without depleting PtdIns(4,5)P₂. We corroborated the previous observations using FRET between PH-PLCδ1 probes (Fig. 7B-D; purple trace). With 1 nM Oxo-M, the development of SOCE was faster and larger than with 1 μM Oxo-M (Fig. 7C,E,F). Adding 100 μM UTP also produced a faster calcium rise (Fig. 7D,E). We conclude that keeping plasma membrane PtdIns(4,5)P₂ high enhances the rate and extent of SOCE evoked by muscarinic and purinergic receptors.

We report that PIP5KI isoforms regulate KCNQ2/3 currents and SOCE entry in distinct ways in tsA201 cells and that RASSF4 is not a full mimic of increased expression of PIP5KI isoforms (Table 1). These issues are discussed one at a time.

The two 5-kinases studied catalyze phosphorylation of PtdIns(4)P to yield PtdIns(4,5)P₂. When probed by PH-PLCδ1 and Tubby_C^R332H domains and by KCNQ channels in single-cell biophysical experiments, the actions of overexpressed PIP5KIβ and PIP5KIγ were fully consistent with an anticipated augmentation of plasma membrane pools of PtdIns(4,5)P₂ (Table 1). The two enzymes localized primarily to the plasma membrane (Fig. 1A,B) (Lacalle et al., 2015; Padrón et al., 2003; Wang et al., 2004). They increased plasma membrane labeling by Tubby_C^R332H (Fig. 1E,F), slowed receptor-induced PtdIns(4,5)P₂ depletion reported by PH-PLCδ1 domains (Fig. 1I,K), reduced the effectiveness of VSP reported by KCNQ channels (Fig. 2A,B), speeded the recovery of KCNQ current after brief VSP activation (Fig. 2A,C) and changed the activation and deactivation gating kinetics of KCNQ channels in manners known to signal an increase in PtdIns(4,5)P₂ at the plasma membrane (Fig. 2D-F). Furthermore, as seen in population experiments with mass spectrometry (Fig. 5A-C), both kinases had the expected reciprocal effects on the total cellular amounts of PtdInsP and PtdInsP₂: they reduced PtdInsP and increased PtdInsP₂. There was the possible exception that, although a significant reduction in PtdInsP was present with PIP5KIβ, the increase in PtdInsP₂ was small and not statistically significant. We believe that this exception is an artifact because cells overexpressing PIP5KIβ were unhealthy and sparse. The apparent toxicity of overexpressed PIP5KIβ might be a result of elevated standing basal calcium influx through PIP5KIβ-opened Orai channels, as reported for constitutively active Orai channels (Park et al., 2009). In photometry, calcium imaging and patch clamp experiments, we could select and study individual cells well labeled with fluorescent PIP5KIβ or other overexpressed proteins. On the other hand, for mass spectrometry experiments we had to average across whole populations of cells in each dish with a range of expression levels and health, thereby probably underestimating phosphoinositide changes compared with the cells examined by photometry, calcium imaging and patch clamp. In short, we speculate that there is a significant increase in total PtdIns(4,5)P₂ in PIP5KIβ-overexpressing cells, but there are too few of them in an unhealthy population to register in mass spectrometry.

RASSF4 is primarily cytosolic (Fig. 1A), whereas its downstream effector Arf6 is found both in endosomes and at the plasma membrane (Brown et al., 2001; Chen et al., 2017) where it could recruit and activate PIP5KIβ (Honda et al., 1999) and PIP5KIγ (Krauss et al., 2003). As for the two PIP5KI enzymes, RASSF4 overexpression reduced total PtdInsP and increased total PtdInsP₂, as shown by mass spectrometry (Fig. 5A-C), consistent with a prior study (Chen et al., 2017). We do not have direct evidence that the increased PtdInsP₂ is actually PtdIns(4,5)P₂ rather than PtdIns(3,4)P₂ or PtdIns(3,5)P₂. Despite a clear increase in the total mass of PtdInsP₂ and PtdInsP₃, RASSF4 overexpression alone did not produce any of the anticipated functional effects on the plasma membrane PtdIns(4,5)P₂ pools, as assayed by single-cell experiments with PH-PLCδ1 and Tubby_C^R332H probes and with KCNQ channels (Figs 1 and 2). Remarkably, the initial receptor-induced translocation of PH-PLCδ1 probes was speeded by RASSF4 rather than being slowed down (Fig. 1J,K). Intracellular labeling by PH-PLCδ1 and Tubby_C^R332H (Fig. 1C-F) suggested that more PtdIns(4,5)P₂ resides on intracellular membranes, consistent with reports that Arf6 activity increases PtdIns(4,5)P₂ in endosomal structures by promoting endosomal recycling of plasma membrane (Brown et al., 2001) and inhibiting dense core vesicle exocytosis, a result that was mimicked by overexpression of PIP5KIα and that could be rescued by adding PIP5KIγ protein to permeable cells (Aikawa and Martin, 2003; Brown et al., 2001).

In contrast to our results with RASSF4 alone, coexpression of RASSF4 with PIP5KIγ increased plasma membrane PtdIns(4,5)P₂, consistent with a prior study (Chen et al., 2017). RASSF4 alone may not be able to increase plasma membrane PtdIns(4,5)P₂ if endogenous PIP5KIγ is already maximally activated or is inaccessible to Arf6. In contrast, RASSF4 coexpressed with PIP5K1β did not cause Tubby to redistribute to the plasma membrane. Interestingly, a high proportion of cells coexpressing RASSF4 and PIP5K1β appeared to have abnormal nuclear morphology (Fig. S1I), an observation that was unique to that expression condition and bears some resemblance to observed invaginations of the endo(sarco)plasmic reticulum into the nucleoplasm of cardiac myocytes and epithelial and muscle cell lines (Lee et al., 2018). It is possible that RASSF4 potentiates PIP5K1β at intracellular sites.

RASSF4 also changed the shape of the plasma membrane, suggestive of an altered relationship with the cytoskeleton. The increase in filopodia and lamellipodia (Fig. S1C-E) might be explained by a previously described role of Arf6 in membrane ruffling (Honda et al., 1999). Arf6 activation has been shown to stimulate plasma membrane protrusions enriched in PtdIns(4,5)P₂, and constitutively active Arf6 generates PtdIns(4,5)P₂-rich vacuoles coated with F-actin (Brown et al., 2001) and initiates formation of F-actin foci on the ventral surface of cells and of actin tails trailing endosomal particles (Schafer et al., 2000).

RASSF4 also increased the number of intracellular bodies labeled with Tubby_C^R332H (Fig. S1F-H). Superimposition of several mid-height confocal slices for each cell suggested a concentration of bodies within the nucleus. Several phosphoinositide kinases have been identified in the nucleus, including PIP5KIα, PIP5KIγi4, PIPKIIα and PIPKIIβ (Tan et al., 2015). PtdIns(4,5)P₂ has been identified on nucleoplasmic calcium storage vesicles that also include PIP4KIIα and PIP4KIIβ and InsP₃ receptors (Yoo et al., 2014). Interestingly, nuclear PIP5KIα has been found to reduce cancer risk by stabilizing p53 via interaction with heat shock proteins. It would be interesting to test whether this protective response contributes to the role of RASSF4 in tumor suppression (Choi et al., 2019).

Thus far, we have explained the actions of RASSF4 as a sequestering of some PtdIns(4,5)P₂ from the plasma membrane to intracellular compartments or particles. Alternatively, RASSF4 might sequester PtdIns(4,5)P₂ into plasma membrane domains that are inaccessible to phospholipase C, KCNQ channels, PH domain probes and Tubby. In that scenario, it could be that the membrane extensions (e.g. ruffles and filopodia) observed with RASSF4 are enriched with PtdIns(4,5)P₂ that is inaccessible to these proteins. The hypothetical plasma membrane domains would have to shelter their PtdIns(4,5)P₂ so well that it does not exchange with the rest of the plasma membrane and does not buffer receptor-activated depletion that might occur there. A hypothesis of lipid rafts is discussed later (Hypothesis 3).

The single-cell results with STIM1 and SOCE were more complex than those with PH-PLCδ1 probes, Tubby and KCNQ channels (Table 1). If the effects on the calcium-entry system were purely a result of potentiation by elevated levels of PtdIns(4,5)P₂, we would expect simple promotion and speeding of STIM1 puncta formation accompanied by an increase in calcium entry (SOCE). Overexpressed PIP5KIβ and RASSF4 did follow that paradigm: there was more PtdInsP₂ seen by mass spectrometry, accompanied by elevated basal levels of STIM1 puncta (Fig. 3B,E), faster formation of puncta after thapsigargin (Fig. 3B,D) and somewhat augmented SOCE (Fig. 4). For these actions, RASSF4 was somewhat more effective than PIP5KIβ. ER–plasma membrane contacts were also more abundant in cells overexpressing RASSF4 (Fig. 3F,G), correlating well with the increased abundance of STIM1 puncta. The effects of PIP5KIγ on STIM1 contrasted with those of PIP5KIβ and RASSF4. Although overexpression of PIP5KIγ elevated plasma membrane PtdIns(4,5)P₂, as well as total PtdInsP₂, it did not change the number of basal STIM1 puncta or SOCE and reduced the number of puncta induced by thapsigargin.

Our experiments with low receptor occupancy (Fig. 7) showed that SOCE is more robust (faster and larger calcium rise) in the absence of much PtdIns(4,5)P₂ hydrolysis, suggesting that maintaining plasma membrane PtdIns(4,5)P₂ levels positively influences the formation of active STIM–Orai complexes. Plasma membrane PtdIns(4,5)P₂ and possibly PtdIns(4)P therefore appear to enhance SOCE although they are not required.

The two kinases and RASSF4 each have their own spectrum of actions, despite increasing PtdIns(4,5)P₂ to a similar extent. Our results with mass spectrometry disproved the hypothesis that the kinases and RASSF4 mediate specific effects by generating PtdIns(4,5)P₂ with distinct fatty acyl chains (Fig. 6). We therefore consider four potential mechanisms for this specificity that could be evaluated in future work.

Around the time that PIP5KI enzymes were cloned, a report invoked the idea of pools of PtdIns(4,5)P₂ in different membranes of the cell to explain the specific functional effects of PIP5KI isoforms (Loijens et al., 1996). They suggested a pool of PtdIns(4,5)P₂ at the plasma membrane accessible to phospholipase C, nuclear pools that turned over separately from cytoplasmic pools and a pool for vesicle fusion with the plasma membrane. Using siRNA against specific PIP5KI isoforms, the short splice variant of PIP5KIγ was found to supply the plasma membrane pool of PtdIns(4,5)P₂ accessible to phospholipase C (Wang et al., 2004). Consistent with this idea, we found that YFP-STIM1 labels different structures in each of our transfection conditions: apparently diffuse in the ER in control and PIP5KIγ cells, at the plasma membrane and perinuclear region in PIP5KIβ cells and at the glass-adhering basal surface in RASSF4 cells.

In this hypothesis, the two kinases and RASSF4 show specificity because the indicator processes studied have different lipid requirements and sample different membranes (see Table 1). KCNQ channels, PH-PLCδ1 probes and Tubby reflect changes in PtdIns(4,5)P₂ with some fidelity, and electrical recording from the channels can report only the PtdIns(4,5)P₂ at the plasma membrane. However, the several sequential steps leading up to SOCE might have individual preferences for PtdIns(4)P, PtdIns(4,5)P₂ or PtdIns(3,4,5)P₃ (Korzeniowski et al., 2009; Nakatsu et al., 2012; Walsh et al., 2010), and do involve both the ER and the plasma membrane. The two kinases and RASSF4 decrease total PtdInsP and increase PtdInsP₂ and PtdInsP₃ but not to the same extent and possibly not in the same membranes.

Several studies have suggested that different PIP5KI isoforms produce subpools of PtdIns(4,5)P₂ within the plasma membrane that are distinct both in localization and function (Calloway et al., 2011; Chakrabarti et al., 2015; Choi et al., 2015; Vasudevan et al., 2009; Wang et al., 2004). They invoke the concept of laterally segregated plasma membrane domains characterized by differences in lipid composition: detergent-resistant raft versus detergent-soluble non-raft domains (Lingwood and Simons, 2010). According to Calloway et al. (2011), overexpression of PIP5KIβ increases PtdIns(4,5)P₂ in both detergent-resistant and detergent-soluble membranes, whereas PIP5KIγ increases PtdIns(4,5)P₂ only in detergent-soluble membrane. They also observed a functional divergence between kinase isoforms on a measure of SOCE: FRET between STIM1 and Orai1 was increased by PIP5KIβ overexpression and reduced by PIP5KIγ overexpression.

Lateral segregation of membranes into functional domains can organize specific lipids, enzymes and effectors into micro-islands acting almost as specialized local complexes (Lingwood and Simons, 2010). We suggest that part of the apparent specificity of kinases and RASSF4 signaling arises from the domains they segregate into. However, rafts are usually considered highly dynamic and more like equilibrium phases that offer no major barrier to exchange of lipids on a time scale of several seconds. If that is a correct view, they would not correspond to the hypothetical ‘inaccessible’ domains that prevent KCNQ channels and PH domains from sampling PtdIns(4,5)P₂ generated by overexpression of RASSF4.

Lipids in lipid rafts are thought to be enriched in saturated and longer hydrocarbon chains and in hydroxylated ceramide backbones (Simons and Sampaio, 2011). However, we detected no change in the distribution of acyl chain length and unsaturation across species of PtdInsP₂ and PtdInsP₃ in response to overexpression of PIP5KIβ, PIP5KIγ or RASSF4 (Fig. 6). PtdIns(4,5)P₂ pools might also be organized by ER–plasma membrane tethering proteins such as extended synaptotagmins, TMEM 16/110 and VAMP proteins, and stabilized by PtdIns(4,5)P₂-corraling septins (Maléth et al., 2014; Manford et al., 2012; Quintana et al., 2015). These structural elements have been shown to affect STIM1 puncta formation and Orai activity, and might interact differently with PIP5KIβ and PIP5KIγ.

PIP5KI enzymes dimerize and interact directly with effectors and modulatory proteins. Thus, in Wnt signaling in zebrafish, two lipid kinases, PIP5K1A and PI4KIIα, form a ternary complex with the cytosolic signaling protein dishevelled that more than doubles their activity (Hu et al., 2015; Qin et al., 2009), changing the metabolic path taken by lipid signals (Olivença et al., 2018). Binding of numerous other signaling proteins to PIP5KI enzymes influences enzyme dimerization and catalytic activity (Lacalle et al., 2015). Homo- and heterodimerization of PIP5KIβ and PIP5KIγ strongly increases enzyme activity. It can be expected that, in our work, overexpression of one kinase would strongly enlarge the pool of its homodimers and heterodimers and deplete the pools of homodimers of the other endogenous kinases. Different dimers might localize differently and mediate different effects. In addition, RASSF4 has downstream effects on cellular signaling that extend beyond PIP5KIs. RASSF4 is thought to activate JNK/c-Jun, p21 and p27 and to inhibit Raf/MEK/ERK, PI3K/mTOR, p53 and GSK3 (De Smedt et al., 2018). We suggest that part of the specificity of the kinases and RASSF4 can be explained by the choices of downstream proteins that they associate with.

In summary, our results underscore the complexity of phosphoinositide metabolism and its nuanced roles in ion channel activity and calcium signaling. PIP5KIβ and PIP5KIγ increase PtdIns(4,5)P₂ and modulate KCNQ2/3 channels to a similar extent, but have divergent effects on STIM1 movement. RASSF4 also increases PtdInsP₂, but KCNQ2/3 channels are unaffected by that increased PtdIns(4,5)P₂ generation, except when RASSF4 and PIP5KIγ are overexpressed together. Through mass spectrometry, we reject the hypothesis that different PIP5KI isoforms produce subspecies of PtdIns(4,5)P₂ with different fatty acid combinations. Rather, the divergence in signaling might be explained by different enzyme–effector interactions or distinct membrane domains of PtdIns(4,5)P₂. Looking at the effects of PtdIns(4,5)P₂ across signaling pathways reveals the microscopic heterogeneity and signaling divergence of this deceptively simple lipid signaling system.

Oxotremorine-M, UTP, thapsigargin, ionomycin, sodium formate, HCl and trimethylsilyl-diazomethane (TMS-DM, 2.0 M in diethyl ether or hexanes) were from Sigma-Aldrich, GSK-7975A was from Aobious and Fura-4F was from Invitrogen. Mass spectrometry-grade methanol, chloroform, dichloromethane and acetonitrile were from Fisher Scientific.

tsA201 cells (Sigma-Aldrich), derived from HEK-293 human embryonic kidney cells and certified by the vendor by short tandem repeat profiling, were cultivated in DMEM (Fisher Scientific) with 10% fetal bovine serum (Sigma-Aldrich) and 2% penicillin/streptomycin (Gibco) and utilized between passage 15 and 50. Transcriptome data for HEK 293T cells suggests all three PIP5KI isoforms and RASSF4 are endogenously expressed in these cells at moderate levels (Sultan et al., 2008). Cells were transfected at 75-85% confluency and plated on poly-L-lysine-coated glass chips. Cells were transfected with Lipofectamine 3000 (Invitrogen) with the following cDNA plasmids: human EGFP-PIP5KIβ, CFP-PIP5KIβ, EGFP-PIP5KIγ_i2 and CFP-PIP5KIγ_i2 (PIP5KIγ_i2 is equivalent to PIP5KIγ90; from Rosa Ana Lacalle and Santos Mañes, Centro Nacional de Biotecnología/Consejo Superior de Investigaciones Científicas, Madrid, Spain; Lacalle et al., 2015); human eCFP-PH-PLCδ1and eYFP-PH-PLCδ1 from K-Ras (provided by Kees Jalink, the Netherlands Cancer Institute, Amsterdam); Tubby_C^R332H-YFP (from Andrew Tinker, University College of London, United Kingdom); Dr-VSP (from Yasushi Okamura, Osaka University, Japan); RASSF4-YFP and RASSF4-mCherry (RASSF4-mCh; from Jen Liou, University of Texas Southwestern, USA); and KCNQ2 and KCNQ3 (from David McKinnon, SUNY Stony Brook, USA).

KCNQ2/3 current was recorded by patch clamp in whole-cell configuration at room temperature with a HEKA EPC 9 amplifier (HEKA Elektronik). The resistance of borosilicate glass pipettes was 3-5 MΩ. Series resistance was ≤10 MΩ and was compensated ≥70%. The bath solution consisted of 160 mM NaCl, 2.5 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, 10 mM HEPES and 8 mM glucose adjusted to pH 7.4 with NaOH. Pipette solution consisted of 175 mM KCl, 5 mM MgCl₂, 5 mM HEPES, 0.1 mM BAPTA K₄, 3 mM Na₂ATP and 0.1 mM Na₃GTP adjusted to pH 7.4 with KOH.

For calcium imaging, cells were loaded with 1 μM Fura-4F (Life Technologies) with 0.05% pluronic F-127 for 20 min at room temperature and washed with 2 mM Ca²⁺ bath solution for 20 min at room temperature before recording. Excitation light was controlled by a TILL Polychrome IV monochromator (BioVision Technologies). Fura-4F images were taken with a Plan Fluor 40×/0.15 objective on a Ti-E microscope (Nikon) coupled to a HQ2 camera (Photometrics) controlled by MetaMorph software (Molecular Devices). We measured the fluorescence emission centered at 540 nm and reported calcium as the ratio of emission with 340 nm excitation over the emission with 380 nm excitation (F₃₄₀/F₃₈₀) for each cell. Three bath solutions were prepared containing 0, 0.5 or 2 mM calcium as follows: 155 mM NaCl, 4.5 mM KCl, 10 mM glucose, 5 mM HEPES plus either 3 mM MgCl₂ and 1 mM EGTA for 0 mM Ca²⁺, 1 mM MgCl₂ and 0.5 mM CaCl₂ for 0.5 mM Ca²⁺ or 1 mM MgCl₂ and 2 mM CaCl₂ for 2 mM Ca²⁺. The three solutions were adjusted to pH 7.4 with NaOH.

Cells plated on #0-glass coverslips were fitted to a recording chamber. TIRF footprints were acquired using a Nikon Eclipse Ti-E microscope equipped with a 60×/1.25 oil-immersion objective and Photometrics QuantEM EMCCD camera (Nikon). Time-series images were taken every 10 s at room temperature and analyzed with ImageJ.

Fluorophores were excited with an argon-ion (CFP, GFP and YFP) and helium-neon (mCherry or RFP) lasers and monitored using a LSM 710 confocal microscope (Zeiss) with a 63×/1.40 oil-immersion objective. For some experiments, images were taken with a LSM 880 confocal microscope (Zeiss) coupled with an Airyscan detector. This system was equipped with 405, 561 and 633 nm lasers and an argon lamp. Images were taken every 5 s at room temperature and analyzed with ImageJ.

FRET was measured as described previously (Falkenburger et al., 2010). Briefly, fluorophores were excited with a ramp of excitation from a monochromator (Polychrome IV, TILL Photonics) and monitored using a Nikon diaphot microcope with a 40×/1.30 oil-immersion objective. Emission light from CFP and YFP was separated by a dichroic mirror, filtered with bandpass filters and detected by photodiodes. The FRET ratio (FRETr) was defined as YFP_C/CFP_C after correction for background and bleed through.

The mass spectrometry protocol differed from our previous studies in that lipid extraction was achieved with chloroform/methanol followed precipitation by trichloroacetic acid (TCA) instead of 1:1 methanol/HCl (Traynor-Kaplan et al., 2017) as detailed below. Adherent cells were washed twice with DPBS (Gibco) and scraped from one 35 mm dish (80-90% of confluence) into 1 ml of DPBS; 100 μl of sample was removed for DNA measurements. Then, cells were transferred to 2 ml Lo-Bind polypropylene tubes (Eppendorf) and centrifuged at 30,000 g for 1 min at 4°C. After aspirating DPBS, we added 0.5 M TCA to the pellet, vortexed for 1 min and incubated on ice for 10 min. Following a centrifugation at 30,000 g for 3 min at 4°C, supernatant was removed. Then, 1 ml of 5% (w/v) TCA containing 10 mM EDTA was added to the pellet and vortexed for 1 min, after which the sample was centrifuged at 30,000 g for 3 min at 4°C and the supernatant removed. We repeated the last step twice, and stored the sample at −80°C. Internal standards of 2 ng PtdIns(3,4,5)P₃, 20 ng PtdIns(4,5)P₂, 20 ng PtdIns(4)P and 50-100 ng of PtdIns were added to the precipitates. The lipid analytical internal standards were ammonium salts of 1-heptadecanoyl-2-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phospho-(1′-myo-inositol-3′,4′,5′-trisphosphate) [17:0, 20:4 PtdIns(3,4,5)P₂], 1-heptadecanoyl-2-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phospho-(1′-myo-inositol-4′,5′-bisphosphate) [17:0, 20:4 PtdIns(4,5)P₂], 1-heptadecanoyl-2-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phospho-(1′-myo-inositol-4′-phosphate) [17:0, 20:4 PtdIns(4)P] and 1-heptadecanoyl-2-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phospho-(1′-myo-inositol) [17:0, 20:4 PI] from Avanti Polar Lipids (LIPID MAPS MS Standards). Lipid extraction was initiated with the addition of 670 μl of ice-cold chloroform–methanol–12.1 N HCl (40:80:1), after which the samples were vortexed for 2 min and allowed to sit on ice for another 10 min. Then, 650 μl of chloroform and 300 μl of 1 N HCl were added to generate two phases. Samples were vortexed and phases were separated by centrifugation at 10,000 g for 2 min; the lower phases were collected in a fresh 2 ml tube. An additional 950 μl of a mixture of chloroform:methanol:1.74 M HCl (v:v:v) was added to the upper phase followed by vortexing and centrifugation at 10,000 g for an additional 2 min. The resultant lower phase was combined with the previously collected lower phase and the combination dried under a nitrogen stream using a Biotage evaporator.

The dried extracts were derivatized (methylated) with TMS-DM and quantified by targeted analysis as described (Traynor-Kaplan et al., 2017). Peak areas were determined with QuanLynx software (Waters). Peak areas were converted to milligrams per 10⁶ cells as described using the 37:4 lipid internal standards and DNA-per-sample measurements. Two approximate assumptions were made. (1) The extraction and detection efficiency of phosphoinositides with each of the different fatty acyl combinations was the same as for the 37:4 standard. (2) Peak areas were proportional to lipid concentration. The latter simplification was supported by finding approximately linear standard curves for the 37:4 lipid species with an intercept near zero measured on a background of typical cell extract.

Means and s.e.m. are reported in the text. Sample size (n) and statistical comparisons are provided in the figure legends. The Student's t-test (one-tailed) was used to compare between two groups and one-way ANOVA was used to compare among three or more groups. Means and statistical tests were calculated considering the result from each cell as an independent data point; each experiment was repeated at least three times. Overexpression conditions were always compared with specific same-day controls. Statistical comparisons were made using GraphPad Prism.

We thank Lea M. Miller for technical assistance, Dr Eamonn J. Dickson for helpful advice, Dr Jongyun Myeong for exploratory experiments, and Drs Dickson, Seung-Ryoung Jung, Duk-Su Koh, Bo Hyun Lee and Myeong for insightful comments on the manuscript.